講演抄録/キーワード |
講演名 |
2020-12-10 14:20
Slide Design Assessment Featuring Visual and Structural Analysis ○Shengzhou Yi(UTokyo)・Junichiro Matsugami(Rubato)・Xueting Wang・Toshihiko Yamasaki(UTokyo) AI2020-3 |
抄録 |
(和) |
(まだ登録されていません) |
(英) |
Appealing design of presentation slides is a great way to make the presentation more attractive and easier to understand. However, the design of slides is difficult to novices and there are limited support systems to help them by evaluating their slides. In this paper, the design problems of the presentation slides are recognized by proposed neural network based on the visual features and the structural features. The created dataset contains 856 slide pairs. For each slide pair, one slide was created by a novice and the other one was improved by the advices from professional consultants. Ten check points with high frequencies were summarized by the consultants, which are set as the prediction targets in this study. For the binary classification of each check point, the class distribution is very imbalanced, since only a small part of samples has the responding design problem. Therefore, recent machine learning methods for addressing class imbalance were applied to prediction and proved to be effective for improving the performance of the proposed model. The proposed neural network can achieve the average accuracies of 80.9% and 80.0% on the balanced and imbalanced dataset, respectively. |
キーワード |
(和) |
/ / / / / / / |
(英) |
Presentation Slide / Design Assessment / Feature Fusion / Class Imbalance / / / / |
文献情報 |
信学技報, vol. 120, no. 281, AI2020-3, pp. 13-18, 2020年12月. |
資料番号 |
AI2020-3 |
発行日 |
2020-12-03 (AI) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
AI2020-3 |
|